Error concealment method

Abstract

A method for concealing an error in a motion picture compression/restoration device is disclosed. In the present invention, a damaged macro block is restored by predicting an optimal motion vector V*x and V*y from all possible motion vectors Vx and Vy within a fixed search range of a previous frame by application of a weighted value function. Accordingly, the damaged macro block is replaced with a macro block from a previous frame obtained by the predicted motion vector. The present invention not only improves the visual effects of a restored image, but also reduces the amount of necessary calculation, allowing realtime compensation.

What is claimed is:  
     
       1. A method of concealing errors in a damaged macro block, comprising: 
       determining a search range of a previous frame based upon a type and mode of a frame and setting an initial value of a motion vector in the search range as an initial search position according to a motion type and coding mode of a frame, wherein an average of motion vectors is set as an initial search position if one of a top or bottom macro block of the damaged macro block has motion vectors in a mode of a P frame;  
       predicting an optimal motion vector V* x  and V* y  from motion vectors V x  and V y  within a search range of a previous frame by applying a weighted value function; and  
       replacing the damaged macro block with a macro block of a previous frame obtained by utilizing the predicted optimal motion vector the V* x  and V* y .  
     
     
       2. A method of  claim 1 , wherein the optimal motion vector V* x  and V* y  is a motion vector which yields a least mean square error of pixel luminances of macro blocks surrounding the damaged macro block and pixel luminances of macro blocks surrounding a macro block in a previous frame. 
     
     
       3. A method of  claim 1 , wherein the optimal motion vector V* x  and V* y  is predicted using the following equation:          (       V   x   *     ,     V   y   *       )     =       argmin       (     Vx   ,   Vy     )     ∈   S                         ∑     t   =   1     3                       D   ij   t          (       V   x     ,     V   y       )                           
       where,                  D   ij   1          (       V   x     ,     V   y       )       =                  ∑     k   =   0       N   -   1                         ∑     m   =   1     W                         W   1          (   m   )       [       x1        (       i   -   m     ,     j   +   k       )       -                                      x   ′        1        (       i   -   m   +     V   x       ,     j   +   k   +     V   y         )          ]   2                               D   ij   2          (       V   x     ,     V   y       )       =                  ∑     k   =   0       N   -   1                         ∑     m   =   1     W                         W   2          (   m   )       [       x2        (       i   +   k     ,     j   -   m       )       -                                      x   ′        2        (       i   +   k   +     V   x       ,     j   -   m   +     V   y         )          ]   2                               D   ij   3          (       V   x     ,     V   y       )       =                  ∑     k   =   0       N   -   1                         ∑     m   =   1     W                         W   3          (   m   )       [       x3        (       i   +   m     ,     j   +   k       )       -                                      x   ′        3        (       i   +   m   +     V   x       ,     j   +   k   +     V   y         )          ]   2                           ∑     m   -   1     W                       W   p          (   m   )         =   1     ,     p   =   1     ,   2   ,   3                   
       where {x 1 , x 2 , x 3 } denote pixel luminances of macro blocks adjacent to the top, sides and bottom of the damaged macro block; {x' 1 , x' 2 , x' 3 } denote pixel luminances in a previous frame corresponding to {x 1 , x 2 , x 3  }; {i,j } denote a row and a column addresses of the damaged block in an image; N denotes a size of a macro block; W(m) denotes a weighted function; and W denotes a number of pixel lines used in a comparison. 
     
     
       4. A method of  claim 3 , when M= 2 , wherein the weighted functions in the top block are represented by 
       
         
             W   1 (1) =E{x 1( i,j −2) x 1( i,j −1)}/ E{x 1( i,j −2) 2 } 
         
       
       
         
             W   1 (2)= E{x 1( i,j −3) x 1( i,j −1)}/ E{x 1( i,j −3) 2 }, and  
         
       
       the other weighed functions in the left and bottom blocks are represented by 
       
         
             W   2 (1)= E{x 2( i −2 ,j ) x 2( i −1 ,j )}/ E{x 2( i −2 ,j ) 2 } 
         
       
       
         
             W   2 (2)= E{x 2( i −3 ,j ) x 2( i −1 ,j )}/ E{x 2( i −3 ,j ) 2 } 
         
       
       
         
             W   3 (1)= E{x 3( i,j+N +1) x 3( i,j+N )}/ E{x 3( i,j+N +1) 2 } 
         
       
       
         
             W   3 (2)= E{x 3 i,j+N +2) x 3( i,j+N}/E{x 3( i,j+N +2) 2 } 
         
       
       where E denotes the mathematical expectation whose value may be estimated by using the sample mean. 
     
     
       5. A method of  claim 1 , wherein the weighted value function is applied with a larger weighted factor towards a boundary of the damaged macro block. 
     
     
       6. A method of  claim 1 , further comprising storing a motion vector of the top macro block of the damaged macro block. 
     
     
       7. A method of  claim 1 , wherein setting an initial value of a motion vector includes using either one of or both a top and bottom macro blocks of the damaged macro block. 
     
     
       8. A method of  claim 1 , wherein if both top and bottom macro blocks of the damaged macro block have motion vectors in a mode of a P frame, an average of two motion vectors is set as an initial search position. 
     
     
       9. A method of  claim 1 , wherein if one of either a top or bottom macro block of the damaged macro block has motion vectors in a mode of a P frame, information of either the top or bottom macro block having the motion vector is set as an initial search position. 
     
     
       10. A method of  claim 1 , wherein if neither top or bottom macro block of the damaged macro block has motion vectors in a mode of a P frame mode, an optimal motion vector is searched starting from a macro block in a previous frame at a position identical to the damaged macro block. 
     
     
       11. A method of  claim 1 , wherein if a top macro block of the damaged macro block has a bidirectional motion vector in a B frame mode, storing a forward motion vector. 
     
     
       12. A method of  claim 1 , wherein if a top and bottom macro block of the damaged macro block have either a forward motion vector or a backward motion vector in a B frame mode, an average of existing motion vectors is set as an initial search position. 
     
     
       13. A method of  claim 1 , wherein if a bottom macro block of the damaged macro block has a bidirectional motion vector and a top macro block of the damaged macro block has either a forward or backward motion vector in a B frame mode, an average of either one of forward or backward motion vectors is set as an initial search position depending upon whether the top macro block has a forward or backward motion vector. 
     
     
       14. A method of  claim 1 , wherein if both top and bottom macro blocks of the damaged macro block have a bidirectional motion vector in a B frame mode, an average of forward motion vectors is set as an initial search position. 
     
     
       15. A method of  claim 1 , wherein if a top macro block of the damaged macro block has a bidirectional motion vector in a B frame mode, an initial search position is set in the same manner as a B frame mode having a forward motion vector. 
     
     
       16. A method of  claim 1 , wherein if neither top or bottom macro blocks of the damaged macro block have motion vectors in a B frame mode, an optimal motion vector is searched starting from a macro block in a previous frame at a position identical to the damaged macro block. 
     
     
       17. A method of concealing errors in a damaged macro block, comprising: 
       setting an initial value of a motion vector as an initial search position according to a type and mode of a frame, using either one of or both a top and bottom macro block of the damaged macro block, wherein  
       an average of motion vectors is set as an initial search position if one of a top or bottom macro block of the damaged macro block has motion vectors in a mode of a P frame, and wherein  
       if both top and bottom macro blocks of the damaged macro block have motion vectors in a mode of a P frame, an average of two motion vectors is set as an initial search position, and wherein  
       if neither top or bottom macro block of the damaged macro block has motion vectors in a mode of a P frame mode, an optimal motion vector is searched starting from a macro block in a previous frame at a position identical to the damaged macro block;  
       predicting an optimal motion vector V* x  and V* y  from motion vectors V x  and V y  within a search range of a previous frame by applying a weighted value function; and  
       replacing the damaged macro block with a macro block of a previous frame obtained by utilizing the predicted optimal motion vector the V* x  and V* y .